Download presentation
Presentation is loading. Please wait.
1
KNOWLEDGE REPRESENTATION
Lecture 2 Knowledge Representation
2
Knowledge Representation: Principles and Techniques
Lexical ambiguity of natural language “Visiting aunts can be a nuisance” Need for common sense knowledge “The hammer hit the vase and it broke” “The vase hit the wall and it broke” “the X hit the Y and it broke” In each of the above, what does ‘it’ refer to ? Lecture 2 Knowledge Representation
3
Predicate-argument Expression
<sentence><predicate>(<argument>,…,<argument>) e.g. at(robot, roomA) Interpretation: Robot is located in Room A (and not Room A is located at the robot) Lecture 2 Knowledge Representation
4
Knowledge Representation
STRIPS Planner push(X,Y,Z) Preconditions: at(robot,Y), at(X,Y) Delete list: at(robot,Y), at(X,Y) Add list: at(robot,Z), at(X,Z) move(X,Y) Preconditions: at(robot,X) Delete list: at(robot,X) Add list: at(robot,Y) Lecture 2 Knowledge Representation
5
STRIPS Planner in Action
Start at(robot,roomA) at(box1,roomB) at(box2,roomC) move(roomA,roomB) X/roomA,Y/roomB P: at(robot,roomA) D: at(robot,roomA) A: at(robot,roomB) push(box1,roomB,roomA) X/box1,Y/roomB,Z/roomA P: at(robot,roomB) P: at(box1,roomB) D: at(robot,roomB) D: at(box1,roomB) A: at(robot,roomA) A: at(box1,roomA) move(roomA,roomC) X/roomA,Y/roomC A: at(robot,roomC) push(box2,roomC,roomA) X/box2,Y/roomC,Z/roomA P: at(robot,roomC) P: at(box2,roomC) D: at(robot,roomC) D: at(box2,roomC) A: at(box2,roomA) Goal at(box1,roomA) at(box2,roomA) Lecture 2 Knowledge Representation
6
Knowledge Representation
MYCIN Domain: treatment of blood infection Decision process: Deciding if patient has significant infection Determining possible organisms involved Selecting a set of appropriate drugs Choosing most appropriate drug(s) Lecture 2 Knowledge Representation
7
Organization of MYCIN Consultation program Explanation program
Knowledge acquisition program Static knowledge base Dynamic patient data Physician user Infectious disease expert Lecture 2 Knowledge Representation
8
MYCIN’s Control Structure
IF 1) there is an organism which requires therapy, and 2) consideration has been given to any other organisms requiring therapy THEN compile a list of possible therapies, and determine the best one in this list. Lecture 2 Knowledge Representation
9
Knowledge Representation
Consultation Session Create the patient context as the top node in the context tree; Attempt to apply the goal rule to this patient context. Consultation is essentially a search through a tree of goals. The leaves of the tree as fact goals, such as laboratory data. Lecture 2 Knowledge Representation
10
An example of MYCIN’s rules
IF 1) The stain of the organism is gramneg, and 2) The morphology of the organism is rod, and 3) The aerobicity of the organism is aerobic THEN There is strongly suggestive evidence (.8) that the class of the organism is enterobacteriaceae Lecture 2 Knowledge Representation
11
Knowledge Representation
MYCIN Context Tree Patient-1 Culture-1 Culture-2 Culture-3 Operation-1 Organism-1 Organism-2 Organism-3 Drug-1 Drug-2 Lecture 2 Knowledge Representation
12
Example data for organism-1
GRAM = (GRAMNEG 1.0) MORPH = (ROD 0.8) (COCCUS 0.2) AIR = (AEROBIC 0.6) Considering the preceding rule, CF(premise) = min(1.0 , 0.8 , 0.6) = 0.6 CF(action) = CF(premise) × CF(rule) = 0.6 × 0.8 = 0.48 Lecture 2 Knowledge Representation
13
Knowledge Representation
AND/OR Tree BADGE SHIELD GUN POLICE REVOLVER PISTOL RIFLE Lecture 2 Knowledge Representation
14
Knowledge Representation
Evidence Combination Let X and Y be weights derived from the application of two different rules. Combined certainty of both rules: Lecture 2 Knowledge Representation
15
Knowledge Representation
Evaluation of MYCIN Ratings by 8 experts on 10 cases Perfect score = 80 MYCIN 52 Actual Therapy 46 Faculty-1 50 Faculty-4 44 Faculty-2 48 Resident 36 Inf dis fellow Faculty-5 34 Faculty-3 Student 24 Lecture 2 Knowledge Representation
16
Comparing MYCIN and STRIPS
Both have goal-driven control structure STRIPS is knowledge-poor (weak method) MYCIN is domain specific (strong method) Lecture 2 Knowledge Representation
17
Symbolic Representation
Syntactic rules to form symbol structures out of symbols Transformation rules to turn symbol structures into other symbol structures Programs in such languages are themselves symbol structures Lecture 2 Knowledge Representation
18
Physical Symbol System Hypothesis
“A physical symbol system has the necessary and sufficient means for general intelligent action” − Newell & Simon (1976) Lecture 2 Knowledge Representation
19
Symbol Structures in LISP
LISP: LISt Processing LISP differs from most programming languages in three important respects: Its main data structure is the list; Its programs are also represented by list structures; and Its primitive operations are operations on lists. Lecture 2 Knowledge Representation
20
Knowledge Representation
Lists and Dotted Pairs Any atom is an S-expression If A1 and A2 are S-expression, then (A1 . A2) is an S-expression If S = (A1 . (A2 . (… . (An-1 . An)…) is an S-expression for n>0, then S is a list if and only if An = NIL (S1 . (S2 . (S3 . NIL))) is simply (S1 S2 S3) (A . B) is not a list. It is a dotted pair. Lecture 2 Knowledge Representation
21
Knowledge Representation
LISP Programs (<function> <1st argument> <2nd argument>) e.g. (+ X Y) To treat a list as merely a data structure rather than a program, use “quote” (quote X) or ’X ’( ( ) ( ) ( ) ) Lecture 2 Knowledge Representation
22
Function Definition in LISP
(defun <function name> (<formal parameters>) <function body> ) e.g. (defun SQUARE (X) (* X X ) ) Lecture 2 Knowledge Representation
23
Processing Association Lists
(defun assoc (key alist) (cond ((null alist) NIL) ((eq (first (first alist)) key) (first alist)) (t (assoc key (rest alist)))) (assoc ’alaska ’( (alabama montgomey) (alaska juneau) …) ) Lecture 2 Knowledge Representation
24
Knowledge Representation
Pattern Matching (defun match (sample pattern) (cond ((and (null sample) (null pattern)) T) ((or (null sample) (null pattern)) NIL) ((eq (first pattern) ’?) (match (rest sample) (rest pattern) ) ) ((eq (first sample) (first pattern) ) (T NIL) ) Lecture 2 Knowledge Representation
25
Pattern Matching (contd.)
(match ’(at robot room) ’(at robot ?) ) T (match ’(at box room) ’(at robot ?) ) NIL (match ’(at robot room) ’(at ? ?) ) Lecture 2 Knowledge Representation
26
Function Definition in CLIPS
(deffunction hypotenuse (?a ?b) (sqrt (+ (* ?a ?a) (* ?b ?b) ) ) ) (hypotenuse 3 4) 5.0 Lecture 2 Knowledge Representation
27
CLIPS for Rule-based Systems
Freeware, available at: Borrows features from other tools Fairly standard LISP-like syntax Reasonably efficient Reasonably flexible Facilities for combining rules with objects Lecture 2 Knowledge Representation
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.